About this corpus

The symphonic experiment corpus exists out of a myriad of symphonies, from a wide variety of composers. What is so interesting about this corpus, is that it spans over a period from the mid 18th century to the present day. This means that the corpus contains a certain chronological storyline, if we were to put all the symphonies side to side. On the other hand, the corpus is not only about the historical composers and their lifes. This music has been played and recorded by actual musicians of the present day (however, assuming that the ‘present day’ is the recording era). From this questions may arise like: from which country are these artists? what artists play only the popular composers? And can we see a trend in quality and diversity of recording? Technical aspects aside, we may also look at the more musical aspects of this corpus: how have the symphonies evolved through time? It is a common conception that contemporary symphonies have ‘strayed away’ from their original, classical, form. Is there actually a scientific proof to debunk or substantiate this? To investigate, among others, the above mentioned questions, this corpus aims to serve as a representation of the symphonic timeline. Typical examples are the Haydn, Mozart and Beethoven symphonies, as they are written in the early classical period and are therefore strongly structered. However one must beware of possible prejudices when compiling such a corpus, because a strong preference for well-known composers may disturb the chronology and variety. That is why I also included some lesser known composers to keep things balanced and diverse. On the other hand, it is merely impossible to preserve a strict chronology in the corpus as musical history capricious; earlier eras such have less composers producing more output, in later eras such as the 19th century, symphonies became more popular and the number of symphonic composers steadily increased. Should we therefore add more romantic symphonies as they are more prominently present in musical history? As this example shows, picking the right symphonies is an arduous labour. Therefore I added a layer of ‘randomness’ to my selection process on the basis of a chronological list of composers. The symphonic experiment corpus only exists out of full symphonies (so works of 3-4 movements or more). As the term ‘classical’ would refer to the classical period, I would rather define the overall genre as ’composed music

Introduction

In the world of classical music, there is a tendency to endlessly re-listen famous classical composers such as Mozart, Bach, Beethoven etc. These composers deserve to have their admirable reputation, however, this has as result that nearly 95 percent of all classical music played in concert halls are old works. To substantiate this claim, in the next visualization, we show the correlation (Pearson’s R) between the popularity of a certain track, and the popularity of the accessory composer. This way it becomes visible how inclined people are to listen to symphonies from famous composers rather than from unknown composers. However the main goal of this research is to find out how symphonies have changed over time and in what way.

Composer vs track


As is visible in the graph, the black line resembles a straight upwards trend. Which means there is a positive correlation between a composers popularity and the popularity of their symphonies. The total Pearson’s R amounts to 0.59 (approximately). People are more inclined to listen to symphonies from famous composers than symphonies from unknown composers.

Discussion

the former visualization shows us that people tend to be more interested in famous composer’s symphonies, than those from unknown composers. From one perspective, this is expected because a composer’s popularity is probably calculated from their tracks’ popularity. But from another perspective would it be somewhat prejudiced to listen to a composer’s symphony purely based on their other works, disregarding the efforts of unknown composers.


Haydn vs. Ades: Ceptograms


To take a closer look at the symphonic corpus, we may consider Joseph Haydn as one of the centre points of classical symphonies and Thomas Ades’ symphonies as an outlier. The symphonies in question are: Symphony no. 44 mvt. I, Haydn and Asyla op. 17 mvt I, Ades. Both representing an extreme within the classical music: Haydn (18th century) being extremely structured and Ades (20th century) being fairly unconventional. These two graphs substantiate that contrast. As is visible, Haydn’s symphony is timbre-wise somewhat restricted and Ades’ symphony’s timbre is more scattered. This comes with no suprise: Haydn restricts himself to a (relatively) small 18th century orchestra, with little or no real variation through the movement. Thomas Ades on the other hand is known for his sudden outbursts, contrastive sections (long silences vs intense soundscapes etc.).

Haydn vs Ades: timbre-based self-similarity matrices


These two matrices respectively substantiate the previous claims. Haydn’s symphonies show little to no timbre variation. Thomas Ades’ however, does show a lot of variety in timbre. Short discussion: How is it possible Ades is less conventional but has more consistent timbre features? A short answer is that modern composers such as Ades often work with smaller/restricted ensembles. Classical composers such as Haydn or Beethoven used big orchestras with a lot of timbre variety. Because Ades’ orchestra is more restricted, his timbre features are more restricted as well.

Haydn vs Ades: chroma-based self-similarity matrices


From these two matrices it becomes very clearly visible how both composers handle structure. If we take a closer look at Haydn’s symphony we see a very clear and obvious structure. The first block (along the diagonal) represents the first material in a short exposition, the second block is the introduction of new material again. Now, in the third and the forth block, we see a literal repetition of these first two blocks. This is an indication of the classical tendency to literally repeat certain sections in a symphony. Thomas Ades is a modern composer. Modern composers are known for their freedom with respect to traditional structures and forms, as can be seen in his matrix. There may be some similarity, but there is no real pattern or structure. For example, in Haydn’s symphony the first blocks are of the same length. What both composers have in common however, is that the patterns from the matrix become blurrier as they move to the final section of their movement.

Histograms


When we plot our corpus in terms of a histogram, it is clearly visible that the most symphonies in our corpus are from the 19th century. It is true that, quantitavely, most symphonies are written between 1780-1820. That may be the cause to the bias in this corpus.

Chordogram


As is visible in this chordogram. No real dominant chord or tonality can be detected. According to the spotify API, the piece has key=0, mode=1, which means the pice should be in C major. However, when listening to this piece, one would not say it were it C major. This is another important feature of how the symphony has progressed through time: complete expressiveness instead of structeral balance.

Haydn Chordogram


In this chordogram of the same piece from Haydn as discussed before, again a clear structure is visible. If we can still remember the exposition from the last time, in which a certain section is repeated once, that is now also harmonically visible. The piece starts in e minor and progresses to G major (common harmonic step III). After repeating that, there is a short section in the sub-dominant A major (IV), which is again, a very common progression. The piece ends again in e minor. Which is expected because it was very uncommon in those times to end a piece in a different key than the key in which it started.

Standard Deviation


This plot shows that on the basis of tempo, duration and volume, already some distinctions can be seen through the century as symphonies flock in groups.

Tempo Histograms


This picture shows the different tempi over the centuries of symphonical tracks. As is visible, the distribution remains somewhat the same over the years.

Tempograms


These are tempograms for two outliers (respectively): a symphony by P. Glass and a symphony by Gorecki. Both composers are known for their contemporary take on the symphony with Glass being a minimalist (repeating the same musical ideas over and over again with little variation). The latter being clearly visible in the tempogram that shows little to no variation in tempi. The same holds for Gorecki.

compare to self similarity


Here are two recordings of the same symphony; symphony no 1 by Prokofiev. The former being made by Bernstein, the last one being made by the ‘dutch national philharmonic orchestra’. As you can see the tracks are slightly different.

compare to self similarity II

Predict the century


one of the main goals of this research is to find chronological stylistic changes in symphonies. Can they be separated by century? And if so, what distinguished them?

As you can see in the results, the random forest algorithm does a pretty okay job in labeling the tracks. precision would typically be around 73-85 and recall would be around 76-85. However this does leave room for enhancements. On the next page we will find out what the most entropic features are.

RESULTS:

# A tibble: 3 x 3
  class precision recall
  <fct>     <dbl>  <dbl>
1 18th      0.810   0.85
2 19th      0.812   0.65
3 20th      0.783   0.9 

Enhancements I


As you can see, the most important features are c05, G, c06 and danceability. Lets feed those to the classifier on the next page

Enhancements II


The results have now been improved by some percents

RESULTS:

# A tibble: 3 x 3
  class precision recall
  <fct>     <dbl>  <dbl>
1 18th      0.810   0.85
2 19th      0.75    0.75
3 20th      0.789   0.75

Conclusion

In conclusion, what can we say about the way symphonies have progressed through time? We have seen that the more classical symphonies have a lot of structure. We can even see that in earlier symphonies, the chromagram and chordograms even match up. Differences in chordograms can be explained through chromagrams, as sections are better visible in chromagrams. We have learned from timbre features that symphonies back in the day had a bigger orchestra, and were therefore more diverse (timbre-wise). This means modern symphonies show less variety in timbre features overall. That may also be the explanation to why the discriminative values of the timbre features are so high in the classification of symphonies. This may even be the most important distiction between symphonies through time.